Morse and Lyapunov Spectra and Dynamics on Flag Bundles

Total Page:16

File Type:pdf, Size:1020Kb

Morse and Lyapunov Spectra and Dynamics on Flag Bundles Morse and Lyapunov Spectra and Dynamics on Flag Bundles∗ Luiz A. B. San Martin† Lucas Seco‡ Instituto de Matem´atica Universidade Estadual de Campinas Cx.Postal 6065, 13.081-970 Campinas-SP, Brasil October 25, 2018 Abstract This paper studies characteristic exponents of flows in relation with the dynamics of flows on flag bundles. The starting point is a flow on a principal bundle with semi-simple group G. Projection against the Iwasawa decomposition G = KAN defines an additive cocycle over the flow with values in a = log A. Its Lyapunov exponents (lim- its along trajectories) and Morse exponents (limits along chains) are studied. It is proved a symmetric property of these spectral sets, namely invariance under the Weyl group. It is proved also that these sets are located in certain Weyl chambers, defined from the dynamics arXiv:0709.3278v1 [math.DS] 20 Sep 2007 on the associated flag bundles. As a special case linear flows on vector bundles are considered. AMS 2000 subject classification: Primary: 37B05, 37B35. Secondary: 20M20, 22E46. Key words: Morse exponents, Lyapunov exponents, Chain transitivity, semi- simple Lie groups, flag manifolds, Morse decomposition. ∗Research supported by FAPESP grant n0 02/10246-2 †Supported by CNPq grant n◦ 305513/03-6 ‡Supported by FAPESP grant n◦ 04/00392-7 1 1 Introduction In this paper we study characteristic exponents (Morse and Lyapunov vector spectra) associated to continuous flows evolving on principal bundles. The purpose is to relate these exponents to the dynamics on the flag bundles. We work with invariant flows on a general principal bundle whose structure group G is semi-simple or reductive. The objects to be considered are de- fined intrinsically from G and the principal bundle. Thus our characteristic exponents take values on a vector subspace a of the Lie algebra g of G, while the flag bundles in question have fiber F, a compact homeneous space of G (generalized flag manifolds of G). When G is a linear group a becomes a subspace of diagonal matrices and F a manifold of flags of subspaces. The vector valued spectra are defined via the cocycle over the flag bun- dle obtained by projection against an Iwasawa decomposition of G. They measure the exponential growth ratio of the flow the same way as the usual one-dimensional spectra measures the growth ratio of linear flows on vector bundles. (Actually the one-dimensional spectrum can be recovered from the vector valued one, via a representation of G, see Section 9.) This is the set up of Kaimanovich [16] to the proof of the multiplicative ergodic theorem of Oseledets. The Morse spectrum set is a concept of growth ratio along chains intro- duced by Colonius-Kliemann [5], [6] (see also Colonius-Fabri-Johnson [7] for a vector valued version). Each one of these sets depend on a chain component of a flow. In our case the Morse spectra are defined over chain components of the flow on an associated flag bundle. Our purpose here is to relate the geometry of the vector Lyapunov and Morse spectra with the geometry of the chain components on the flag bundle, which were studied in [4] and [22]. We better explain our results with the concrete example where G = Sl(3, R). Let X be a compact Hausdorff space endowed with a continu- ous flow (t, x) 7→ t · x (t ∈ T = Z or R) and suppose that φt (x, g) = (t · x, ρ (t, x) g) is a flow on Q = X × G with ρ a continuous cocycle taking values in G. Let A ⊂ G be the subgroup of diagonal matrices (positive en- tries) and N the subgroup of upper triangular unipotent matrices. Then the Iwasawa decomposition of G reads G = KAN with K = SO(3). For x ∈ X and u ∈ K we write ρ (t, x) u = kt (x, u) at (x, u) nt (x, u) ∈ KAN and look at the limit behavior as t →∞ of the matrix a (t, x, u) /t = log at (x, u) /t in the subspace a = log A. 2 This is done in combination with the asymptotics of the compact part kt (x, u). Via the action of G on K ≃ G/AN the component kt (x, u) becomes a flow on K and a (t, x, k) an additive cocycle over this flow. Thus the characteristic exponents defined by a depend on (x, k) ∈ X ×K. Nevertheless it is more convenient to avoid ambiguities and factor out the subgroup M ⊂ K of diagonal matrices with ±1 entries and get a (t, x, b) as a cocycle over the flow φF induced on X × F, where F ≃ K/M ≃ G/MAN is the manifold of complete flags in R3. Now let M be a chain component of the flow φF on X × F. Its associ- ated Morse spectrum set ΛMo (M) ⊂ a is defined by evaluating the cocycle a (t, x, b) on chains in M taking into account the jumps of the chains (see Definition 3.1 below). The Lyapunov spectrum set ΛLy (M) is the set of limits lim a (t, x, b) /t along the trajectories in M. By general facts ΛMo (M) is a compact convex set which contains ΛLy (M), which in turn contains the set of extremal points of ΛMo (M) (see Section 3 below). The chain components in X × F were described in [4], [22] with the as- sumption that the flow on X is chain transitive. The picture is the following: −1 There exists Hφ ∈ a and a continuous map x ∈ X 7→ fφ (x) = gxHφgx , gx ∈ G, into the set of conjugates of Hφ such that each chain component M has the form x∈X {x} × fix (fφ (x)). Here fix(fφ (x)) is a connected com- ponent of the fixed point set of the action on F of exp fφ (x) (the connected components canS be labelled accordingly so that they match each other). For example if the diagonal matrix Hφ has distinct eigenvalues λ1 > λ2 > λ3 with eigenvectors e1, e2, e3 then there are 3! = 6 isolated fixed points, namely the flag b0 = (span{e1} ⊂ span{e1, e2}) together with those obtained from b0 by permuting the basic vectors. A similar description holds for fφ (x), x ∈ X, allowing to label the fixed points, as well as the chain components, with the permutation group on three letters S3, which is the Weyl group W for Sl(3, R). In this case there are 6 distinct chain components denoted by Mw with w running over W = S3. The chain components are also labelled by the permutation group even if Hφ has repeated eigenvalues, but now some of the Mw can merge. In our Sl(3, R) example there are four possibilities for Hφ, namely (a) Hφ = diag(λ1 >λ2 >λ3) with six chain components Mw, w ∈ W. (b) Hφ = diag(λ1 = λ2 > λ3) with three chain components M1 = M(12), M(23) = M(123) and M(132) = M(13). 3 (c) Hφ = diag(λ1 > λ2 = λ3) with three chain components M1 = M(23), M(12) = M(132) and M(123) = M(13). (d) Hφ = diag(λ1 = λ2 = λ3) = 0 and the flow is chain transitive on X × F. (In all the cases M1 is the only attractor component and M(13) the only repeller. The equalities in the second and third cases are due to the cosets Wφw where Wφ is the subgroup of W fixing Hφ.) We say that the matrix Hφ is the block form of φ. It is determined by the dynamics on the flag bundles and is a combination of the parabolic type of the flow and of the reversed flow (see [4], [22] and Section 5 of the present article). The results of this paper provide a similar picture for the Morse and Lyapunov spectra over the chain components in terms of the Weyl group, the Weyl chambers and the block form Hφ of φ. For G = Sl(3, R) the roots of a are the functionals αij (diag(a1, a2, a3)) = ai − aj, for i =6 j ∈{1, 2, 3}. Their kernels are three straight lines containing the nonregular matrices in a, i.e., those having repeated eigenvalues. The complement to the lines is the union of 3! = 6 Weyl chambers where the Weyl group acts simply and transitively (see Figure 1). A chamber is determined by inequalities ai > aj > ak and we denote it by Cw if w ∈ W is the permutation sending (123) to (ijk). The main results of this paper prove the following estimates and symme- try properties of the vector valued Morse and Lyapunov spectra (see Figures 1 and 2): 1. The union w∈W ΛMo (Mw) of the several Morse spectra is invariant under the Weyl group. The same holds for the Lyapunov spectra S w∈W ΛLy (Mw). 2. TheS spectra ΛMo (M1) and ΛLy (M1) of the attractor component are invariant under the subgroup Wφ fixing Hφ. 3. Λ (M ) is contained in the interior of clC . This means that Mo 1 w∈Wφ w the multiplicities of the eigenvalues of any Morse exponent in Λ (M ) S Mo 1 do not exceed those of Hφ. The same estimate holds for ΛLy (M1), since ΛMo (Mw) is the convex closure of ΛLy (Mw). 4. ΛMo (M1) intercepts the closure of every chamber Cw, w ∈ Wφ. Hence there exists a Morse exponent in ΛMo (M1) whose eigenvalues have 4 Figure 1: Morse spectra for flows with G = Sl(3, R): block forms (a) and (b), each spectrum is labeled by its corresponding component.
Recommended publications
  • The Generalized Triangular Decomposition
    MATHEMATICS OF COMPUTATION Volume 77, Number 262, April 2008, Pages 1037–1056 S 0025-5718(07)02014-5 Article electronically published on October 1, 2007 THE GENERALIZED TRIANGULAR DECOMPOSITION YI JIANG, WILLIAM W. HAGER, AND JIAN LI Abstract. Given a complex matrix H, we consider the decomposition H = QRP∗,whereR is upper triangular and Q and P have orthonormal columns. Special instances of this decomposition include the singular value decompo- sition (SVD) and the Schur decomposition where R is an upper triangular matrix with the eigenvalues of H on the diagonal. We show that any diag- onal for R can be achieved that satisfies Weyl’s multiplicative majorization conditions: k k K K |ri|≤ σi, 1 ≤ k<K, |ri| = σi, i=1 i=1 i=1 i=1 where K is the rank of H, σi is the i-th largest singular value of H,andri is the i-th largest (in magnitude) diagonal element of R. Given a vector r which satisfies Weyl’s conditions, we call the decomposition H = QRP∗,whereR is upper triangular with prescribed diagonal r, the generalized triangular decom- position (GTD). A direct (nonrecursive) algorithm is developed for computing the GTD. This algorithm starts with the SVD and applies a series of permu- tations and Givens rotations to obtain the GTD. The numerical stability of the GTD update step is established. The GTD can be used to optimize the power utilization of a communication channel, while taking into account qual- ity of service requirements for subchannels. Another application of the GTD is to inverse eigenvalue problems where the goal is to construct matrices with prescribed eigenvalues and singular values.
    [Show full text]
  • CSE 275 Matrix Computation
    CSE 275 Matrix Computation Ming-Hsuan Yang Electrical Engineering and Computer Science University of California at Merced Merced, CA 95344 http://faculty.ucmerced.edu/mhyang Lecture 13 1 / 22 Overview Eigenvalue problem Schur decomposition Eigenvalue algorithms 2 / 22 Reading Chapter 24 of Numerical Linear Algebra by Llyod Trefethen and David Bau Chapter 7 of Matrix Computations by Gene Golub and Charles Van Loan 3 / 22 Eigenvalues and eigenvectors Let A 2 Cm×m be a square matrix, a nonzero x 2 Cm is an eigenvector of A, and λ 2 C is its corresponding eigenvalue if Ax = λx Idea: the action of a matrix A on a subspace S 2 Cm may sometimes mimic scalar multiplication When it happens, the special subspace S is called an eigenspace, and any nonzero x 2 S is an eigenvector The set of all eigenvalues of a matrix A is the spectrum of A, a subset of C denoted by Λ(A) 4 / 22 Eigenvalues and eigenvectors (cont'd) Ax = λx Algorithmically: simplify solutions of certain problems by reducing a coupled system to a collection of scalar problems Physically: give insight into the behavior of evolving systems governed by linear equations, e.g., resonance (of musical instruments when struck or plucked or bowed), stability (of fluid flows with small perturbations) 5 / 22 Eigendecomposition An eigendecomposition (eigenvalue decomposition) of a square matrix A is a factorization A = X ΛX −1 where X is a nonsingular and Λ is diagonal Equivalently, AX = X Λ 2 3 λ1 6 λ2 7 A x x ··· x = x x ··· x 6 7 1 2 m 1 2 m 6 .
    [Show full text]
  • The Schur Decomposition Week 5 UCSB 2014
    Math 108B Professor: Padraic Bartlett Lecture 5: The Schur Decomposition Week 5 UCSB 2014 Repeatedly through the past three weeks, we have taken some matrix A and written A in the form A = UBU −1; where B was a diagonal matrix, and U was a change-of-basis matrix. However, on HW #2, we saw that this was not always possible: in particular, you proved 1 1 in problem 4 that for the matrix A = , there was no possible basis under which A 0 1 would become a diagonal matrix: i.e. you proved that there was no diagonal matrix D and basis B = f(b11; b21); (b12; b22)g such that b b b b −1 A = 11 12 · D · 11 12 : b21 b22 b21 b22 This is a bit of a shame, because diagonal matrices (for reasons discussed earlier) are pretty fantastic: they're easy to raise to large powers and calculate determinants of, and it would have been nice if every linear transformation was diagonal in some basis. So: what now? Do we simply assume that some matrices cannot be written in a \nice" form in any basis, and that we should assume that operations like matrix exponentiation and finding determinants is going to just be awful in many situations? The answer, as you may have guessed by the fact that these notes have more pages after this one, is no! In particular, while diagonalization1 might not always be possible, there is something fairly close that is - the Schur decomposition. Our goal for this week is to prove this, and study its applications.
    [Show full text]
  • Finite-Dimensional Spectral Theory Part I: from Cn to the Schur Decomposition
    Finite-dimensional spectral theory part I: from Cn to the Schur decomposition Ed Bueler MATH 617 Functional Analysis Spring 2020 Ed Bueler (MATH 617) Finite-dimensional spectral theory Spring 2020 1 / 26 linear algebra versus functional analysis these slides are about linear algebra, i.e. vector spaces of finite dimension, and linear operators on those spaces, i.e. matrices one definition of functional analysis might be: “rigorous extension of linear algebra to 1-dimensional topological vector spaces” ◦ it is important to understand the finite-dimensional case! the goal of these part I slides is to prove the Schur decomposition and the spectral theorem for matrices good references for these slides: ◦ L. Trefethen & D. Bau, Numerical Linear Algebra, SIAM Press 1997 ◦ G. Strang, Introduction to Linear Algebra, 5th ed., Wellesley-Cambridge Press, 2016 ◦ G. Golub & C. van Loan, Matrix Computations, 4th ed., Johns Hopkins University Press 2013 Ed Bueler (MATH 617) Finite-dimensional spectral theory Spring 2020 2 / 26 the spectrum of a matrix the spectrum σ(A) of a square matrix A is its set of eigenvalues ◦ reminder later about the definition of eigenvalues ◦ σ(A) is a subset of the complex plane C ◦ the plural of spectrum is spectra; the adjectival is spectral graphing σ(A) gives the matrix a personality ◦ example below: random, nonsymmetric, real 20 × 20 matrix 6 4 2 >> A = randn(20,20); ) >> lam = eig(A); λ 0 >> plot(real(lam),imag(lam),’o’) Im( >> grid on >> xlabel(’Re(\lambda)’) -2 >> ylabel(’Im(\lambda)’) -4 -6 -6 -4 -2 0 2 4 6 Re(λ) Ed Bueler (MATH 617) Finite-dimensional spectral theory Spring 2020 3 / 26 Cn is an inner product space we use complex numbers C from now on ◦ why? because eigenvalues can be complex even for a real matrix ◦ recall: if α = x + iy 2 C then α = x − iy is the conjugate let Cn be the space of (column) vectors with complex entries: 2v13 = 6 .
    [Show full text]
  • The Non–Symmetric Eigenvalue Problem
    Chapter 4 of Calculus++: The Non{symmetric Eigenvalue Problem by Eric A Carlen Professor of Mathematics Georgia Tech c 2003 by the author, all rights reserved 1-1 Table of Contents Overview ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: 1-3 Section 1: Schur factorization 1.1 The non{symmetric eigenvalue problem :::::::::::::::::::::::::::::::::::::: 1-4 1.2 What is the Schur factorization? ::::::::::::::::::::::::::::::::::::::::::::: 1-4 1.3 The 2 × 2 case ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: 1-5 Section 2: Complex eigenvectors and the geometry of Cn 2.1 Why get complicated? ::::::::::::::::::::::::::::::::::::::::::::::::::::::: 1-8 2.2 Algebra and geometry in Cn ::::::::::::::::::::::::::::::::::::::::::::::::: 1-8 2.3 Unitary matrices :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: 1-11 2.4 Schur factorization in general ::::::::::::::::::::::::::::::::::::::::::::::: 1-11 Section: 3 Householder reflections 3.1 Reflection matrices ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::1-15 3.2 The n × n case ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::1-17 3.3 Householder reflection matrices and the QR factorization :::::::::::::::::::: 1-19 3.4 The complex case ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: 1-23 Section: 4 The QR iteration 4.1 What the QR iteration is ::::::::::::::::::::::::::::::::::::::::::::::::::: 1-25 4.2 When and how QR iteration works :::::::::::::::::::::::::::::::::::::::::: 1-28 4.3 What to do when QR iteration
    [Show full text]
  • Parallel Eigenvalue Reordering in Real Schur Forms∗
    Parallel eigenvalue reordering in real Schur forms∗ R. Granat,y B. K˚agstr¨om,z and D. Kressnerx May 12, 2008 Abstract A parallel algorithm for reordering the eigenvalues in the real Schur form of a matrix is pre- sented and discussed. Our novel approach adopts computational windows and delays multiple outside-window updates until each window has been completely reordered locally. By using multiple concurrent windows the parallel algorithm has a high level of concurrency, and most work is level 3 BLAS operations. The presented algorithm is also extended to the generalized real Schur form. Experimental results for ScaLAPACK-style Fortran 77 implementations on a Linux cluster confirm the efficiency and scalability of our algorithms in terms of more than 16 times of parallel speedup using 64 processor for large scale problems. Even on a single processor our implementation is demonstrated to perform significantly better compared to the state-of-the-art serial implementation. Keywords: Parallel algorithms, eigenvalue problems, invariant subspaces, direct reorder- ing, Sylvester matrix equations, condition number estimates 1 Introduction The solution of large-scale matrix eigenvalue problems represents a frequent task in scientific computing. For example, the asymptotic behavior of a linear or linearized dynamical system is determined by the right-most eigenvalue of the system matrix. Despite the advance of iterative methods { such as Arnoldi and Jacobi-Davidson algorithms [3] { there are problems where a transformation method { usually the QR algorithm [14] { is preferred, even in a large-scale setting. In the example quoted above, an iterative method may fail to detect the right-most eigenvalue and, in the worst case, misleadingly predict stability even though the system is unstable [33].
    [Show full text]
  • The Unsymmetric Eigenvalue Problem
    Jim Lambers CME 335 Spring Quarter 2010-11 Lecture 4 Supplemental Notes The Unsymmetric Eigenvalue Problem Properties and Decompositions Let A be an n × n matrix. A nonzero vector x is called an eigenvector of A if there exists a scalar λ such that Ax = λx: The scalar λ is called an eigenvalue of A, and we say that x is an eigenvector of A corresponding to λ. We see that an eigenvector of A is a vector for which matrix-vector multiplication with A is equivalent to scalar multiplication by λ. We say that a nonzero vector y is a left eigenvector of A if there exists a scalar λ such that λyH = yH A: The superscript H refers to the Hermitian transpose, which includes transposition and complex conjugation. That is, for any matrix A, AH = AT . An eigenvector of A, as defined above, is sometimes called a right eigenvector of A, to distinguish from a left eigenvector. It can be seen that if y is a left eigenvector of A with eigenvalue λ, then y is also a right eigenvector of AH , with eigenvalue λ. Because x is nonzero, it follows that if x is an eigenvector of A, then the matrix A − λI is singular, where λ is the corresponding eigenvalue. Therefore, λ satisfies the equation det(A − λI) = 0: The expression det(A−λI) is a polynomial of degree n in λ, and therefore is called the characteristic polynomial of A (eigenvalues are sometimes called characteristic values). It follows from the fact that the eigenvalues of A are the roots of the characteristic polynomial that A has n eigenvalues, which can repeat, and can also be complex, even if A is real.
    [Show full text]
  • Schur Triangularization and the Spectral Decomposition(S)
    Advanced Linear Algebra – Week 7 Schur Triangularization and the Spectral Decomposition(s) This week we will learn about: • Schur triangularization, • The Cayley–Hamilton theorem, • Normal matrices, and • The real and complex spectral decompositions. Extra reading and watching: • Section 2.1 in the textbook • Lecture videos 25, 26, 27, 28, and 29 on YouTube • Schur decomposition at Wikipedia • Normal matrix at Wikipedia • Spectral theorem at Wikipedia Extra textbook problems: ? 2.1.1, 2.1.2, 2.1.5 ?? 2.1.3, 2.1.4, 2.1.6, 2.1.7, 2.1.9, 2.1.17, 2.1.19 ??? 2.1.8, 2.1.11, 2.1.12, 2.1.18, 2.1.21 A 2.1.22, 2.1.26 1 Advanced Linear Algebra – Week 7 2 We’re now going to start looking at matrix decompositions, which are ways of writing down a matrix as a product of (hopefully simpler!) matrices. For example, we learned about diagonalization at the end of introductory linear algebra, which said that... While diagonalization let us do great things with certain matrices, it also raises some new questions: Over the next few weeks, we will thoroughly investigate these types of questions, starting with this one: Advanced Linear Algebra – Week 7 3 Schur Triangularization We know that we cannot hope in general to get a diagonal matrix via unitary similarity (since not every matrix is diagonalizable via any similarity). However, the following theorem says that we can get partway there and always get an upper triangular matrix. Theorem 7.1 — Schur Triangularization Suppose A ∈ Mn(C).
    [Show full text]
  • Introduction to QR-Method SF2524 - Matrix Computations for Large-Scale Systems
    Introduction to QR-method SF2524 - Matrix Computations for Large-scale Systems Introduction to QR-method 1 / 17 Inverse iteration I Computes eigenvalue closest to a target Rayleigh Quotient Iteration I Computes one eigenvalue with a starting guess Arnoldi method for eigenvalue problems I Computes extreme eigenvalue Now: QR-method Compute all eigenvalues Suitable for dense problems Small matrices in relation to other algorithms So far we have in the course learned about... Methods suitable for large sparse matrices Power method I Computes largest eigenvalue Introduction to QR-method 2 / 17 Rayleigh Quotient Iteration I Computes one eigenvalue with a starting guess Arnoldi method for eigenvalue problems I Computes extreme eigenvalue Now: QR-method Compute all eigenvalues Suitable for dense problems Small matrices in relation to other algorithms So far we have in the course learned about... Methods suitable for large sparse matrices Power method I Computes largest eigenvalue Inverse iteration I Computes eigenvalue closest to a target Introduction to QR-method 2 / 17 Arnoldi method for eigenvalue problems I Computes extreme eigenvalue Now: QR-method Compute all eigenvalues Suitable for dense problems Small matrices in relation to other algorithms So far we have in the course learned about... Methods suitable for large sparse matrices Power method I Computes largest eigenvalue Inverse iteration I Computes eigenvalue closest to a target Rayleigh Quotient Iteration I Computes one eigenvalue with a starting guess Introduction to QR-method 2 / 17 Now: QR-method Compute all eigenvalues Suitable for dense problems Small matrices in relation to other algorithms So far we have in the course learned about..
    [Show full text]
  • 10. Schur Decomposition
    L. Vandenberghe ECE133B (Spring 2020) 10. Schur decomposition • eigenvalues of nonsymmetric matrix • Schur decomposition 10.1 Eigenvalues and eigenvectors a nonzero vector x is an eigenvector of the n × n matrix A, with eigenvalue λ, if Ax = λx • the eigenvalues are the roots of the characteristic polynomial det¹λI − Aº = 0 • eigenvectors are nonzero vectors in the nullspace of λI − A for most of the lecture, we assume that A is a complex n × n matrix Schur decomposition 10.2 Linear indepence of eigenvectors suppose x1,..., xk are eigenvectors for k different eigenvalues: Ax1 = λ1x1;:::; Axk = λk xk then x1,..., xk are linearly independent • the result holds for k = 1 because eigenvectors are nonzero • suppose it holds for k − 1, and assume α1x1 + ··· + αk xk = 0; then 0 = A¹α1x1 + α2x2 + ··· + αk xkº = α1λ1x1 + α2λ2x2 + ··· + αkλk xk • subtracting λ1¹α1x1 + ··· + αk xkº = 0 shows that α2¹λ2 − λ1ºx2 + ··· + αk¹λk − λ1ºxk = 0 • since x2,..., xk are linearly independent, α2 = ··· = αk = 0 • hence α1 = ··· = αk = 0, so x1,..., xk are linearly independent Schur decomposition 10.3 Multiplicity of eigenvalues Algebraic multiplicity • the multiplicity of the eigenvalue as a root of the characteristic polynomial • the sum of the algebraic multiplicities of the eigenvalues of an n × n matrix is n Geometric multiplicity • the geometric multiplicity is the dimension of null¹λI − Aº • the maximum number of linearly independent eigenvectors with eigenvalue λ • sum is the maximum number of linearly independent eigenvectors of the matrix Defective eigenvalue
    [Show full text]
  • Section 6.1.2: the Schur Decomposition 6.1.3: Nonunitary Decompositions
    Section 6.1.2: The Schur Decomposition 6.1.3: Nonunitary Decompositions Jim Lambers March 22, 2021 • Now that we have answered the first of the two questions posed earlier, by identifying a class of matrices for which eigenvalues are easily obtained, we move to the second question: how can a general square matrix A be reduced to (block) triangular form, in such a way that its eigenvalues are preserved? • That is, we need to find a similarity transformation (that is, A P −1AP for some P ) that can bring A closer to triangular form, in some sense. n n×n • With this goal in mind, suppose that x 2 C is a normalized eigenvector of A 2 C (that is, kxk2 = 1), with eigenvalue λ. • Furthermore, suppose that P is a Householder reflection such that P x = e1. (this can be done based on Section 4.2) • Because P is complex, it is Hermitian, meaning that P H = P , and unitary, meaning that P H P = I. • This the generalization of a real Householder reflection being symmetric and orthogonal. • Therefore, P is its own inverse, so P e1 = x. • It follows that the matrix P H AP , which is a similarity transformation of A, satisfies H H H P AP e1 = P Ax = λP x = λP x = λe1: H H • That is, e1 is an eigenvector of P AP with eigenvalue λ, and therefore P AP has the block structure λ vH P H AP = : (1) 0 B • Therefore, λ(A) = fλg [ λ(B), which means that we can now focus on the (n − 1) × (n − 1) matrix B to find the rest of the eigenvalues of A.
    [Show full text]
  • Comparison of Numerical Methods and Open-Source Libraries for Eigenvalue Analysis of Large-Scale Power Systems
    applied sciences Article Comparison of Numerical Methods and Open-Source Libraries for Eigenvalue Analysis of Large-Scale Power Systems Georgios Tzounas , Ioannis Dassios * , Muyang Liu and Federico Milano School of Electrical and Electronic Engineering, University College Dublin, Belfield, Dublin 4, Ireland; [email protected] (G.T.); [email protected] (M.L.); [email protected] (F.M.) * Correspondence: [email protected] Received: 30 September 2020; Accepted: 24 October 2020; Published: 28 October 2020 Abstract: This paper discusses the numerical solution of the generalized non-Hermitian eigenvalue problem. It provides a comprehensive comparison of existing algorithms, as well as of available free and open-source software tools, which are suitable for the solution of the eigenvalue problems that arise in the stability analysis of electric power systems. The paper focuses, in particular, on methods and software libraries that are able to handle the large-scale, non-symmetric matrices that arise in power system eigenvalue problems. These kinds of eigenvalue problems are particularly difficult for most numerical methods to handle. Thus, a review and fair comparison of existing algorithms and software tools is a valuable contribution for researchers and practitioners that are interested in power system dynamic analysis. The scalability and performance of the algorithms and libraries are duly discussed through case studies based on real-world electrical power networks. These are a model of the All-Island Irish Transmission System with 8640 variables; and, a model of the European Network of Transmission System Operators for Electricity, with 146,164 variables. Keywords: eigenvalue analysis; large non-Hermitian matrices; numerical methods; open-source libraries 1.
    [Show full text]